The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a PreToolUse hook for the Bash tool. This hook executes a local script guard-ship-irreversibles.sh to monitor and intercept potentially destructive Git operations. This is a security-enhancing feature designed to enforce human-in-the-loop requirements for irreversible actions.\n- [REMOTE_CODE_EXECUTION]: The skill uses python3 to parse and validate shell commands within its guard script. The Python logic is embedded directly in the script (via python3 -c) and operates on local environment data. No remote scripts are downloaded or executed.\n- [EXTERNAL_DOWNLOADS]: The skill does not perform any network operations or downloads from external servers. All operations are local to the project environment and the provided script files.\n- [PROMPT_INJECTION]: The instructions contain strong 'Non-negotiable' directives and 'Extremely Important' sections. These are used to reinforce safety boundaries (e.g., 'GATES FAIL CLOSED', 'HUMAN SIGN-OFF') rather than bypassing them. The adversarial reasoning section ('Common Rationalizations') is designed to prevent the model from talking itself out of safety protocols.

auto-ship